The IEA's 2030 number is 950 terawatt-hours, and roughly half the growth has no committed generation yet

950 terawatt-hours is the number that made the headlines, and it is the less interesting half of the projection.

The demand half first, because it is genuinely large. The IEA's 2026 Electricity report projects global data-center power demand growing from 485 TWh in 2025 to 950 TWh by 2030. At that level, data centers alone would draw roughly 4 percent of all electricity consumed on earth. AI-focused data centers grow threefold across the same five years. The regional splits are steep everywhere: the United States adds 240 TWh, a 130 percent increase; China adds 175 TWh, a 170 percent increase; Europe adds 45 TWh, a 70 percent increase.

Those are demand numbers. Demand numbers are the easy half of any forecast, because demand is the thing buyers are already saying out loud. Hyperscaler capex guidance and the price of a two-generation-old GPU say the same thing, in public, with money behind it. The supply half of the same report is quieter, and it is the half that decides whether any given developer's 2030 actually happens.

The supply half is where the report gets honest

The IEA projects renewables meeting roughly half of this demand growth, with natural gas adding roughly 175 TWh worth of generation to help cover the rest. Run the subtraction. Growth from 485 to 950 TWh is 465 TWh of new annual demand by 2030. Roughly half of that is assigned, on paper, to renewable buildout that mostly has not been built, and in plenty of cases has not even been proposed. A gas tranche of roughly 175 TWh covers much of the remainder, and those plants are largely prospective too. The word "projected" is carrying an enormous load in both clauses, and the balance between them is exactly the part no forecaster controls.

A projection is not a power-purchase agreement.

That sentence is the whole post, so let me spend the rest of it on what the gap means for one developer with one site, because in the aggregate the numbers look like an energy-policy debate, and at the parcel level they look like your term sheet.

Say you are locking a site in the second half of 2026 for energization in 2029 or 2030, which is the realistic window for anything starting diligence today. The load you are bringing sits inside that 465 TWh of growth. The electrons you need sit inside a supply plan of which roughly half is a committed-in-spirit renewable trajectory and the rest is gas that has to clear its own siting and permitting before it generates anything. Your counterparty for a power-purchase agreement or an interconnection commitment is pricing that same uncertainty from the other side of the table. The utility does not know which of its queued generation projects will be operating in 2030 any more than you know which of your entitlement hearings will run long. Both of you are underwriting a stack of public processes neither of you controls.

Which is why the uncommitted half should change your diligence behavior in a specific way. The question stops being "what is the market's 2030 demand number" and becomes "which serving utilities have committed generation and interconnection queues shallow enough that my 2029 date survives contact with reality." That answer varies by jurisdiction more than any national number suggests. The US regional figure, 240 TWh of added demand against a 130 percent load increase, is not distributed evenly; it lands wherever the queues clear, and it walks away from wherever they do not. The power bill is the pro forma was our phrase for this before the IEA put a global number on it. The resource plan deserves the same hours of reading a zoning code gets. It is the same document in a different costume: a public record of what a jurisdiction will actually let you build, and when.

The regional numbers also describe different kinds of scarcity, which is easy to miss when they sit in one table. China's 170 percent growth runs through a state-directed grid, where a generation commitment and a construction start sit unusually close together. Europe's 70 percent runs through permitting regimes nobody has ever accused of speed, which is part of why its absolute number is the smallest of the three. The American 130 percent runs through dozens of state commissions and thousands of local boards, the most fragmented approval system on the list, wrapped around the largest absolute prize. Same forecast, three completely different underwriting problems.

Nobody making a 2030 promise today controls the generation behind it.

The practical posture, then. Treat every generation forecast in a utility conversation the way you treat an unsigned term sheet, as a mood rather than a commitment. Ask which specific plants back your load and where your interconnection request sits in the queue, by position number rather than by adjective. If the answers are a forecast, a plan, and "soon," price the uncertainty into the land basis, because someone will, and it is better if it is you.

My read is that the demand half of the IEA's projection is the safe half. I expect global data-center demand to arrive at or above 950 TWh by 2030, and I expect the interesting misses to be regional and supply-side: markets that wanted the load and could not energize it, watching it land one state or one border away. If 2030 arrives with demand under 950 and generation to spare, I had this exactly backwards and will say so. Until then, the developers I would back are the ones reading the supply half of the report, because 465 TWh of growth is only an opportunity for the sites that can plug in.

This analysis is a source-cited research summary drawn from public records, not legal advice. It can contain errors and should be verified independently before any investment decision.